Predicting Marine Fuel with High Sulphur Content Using Machine Learning Algorithms

Abstract

Marine transportation is a significant source of air pollution especially around coastal areas with maritime vessels creating 12% of global sulphur oxides emission in 2014 alone. In compliance with International Maritime Organisation (IMO) regulations, the determination of sulphur content of marine fuels is typically carried out using lengthy laboratory-based analyses. The regulations prohibit the use of High-Sulphur Fuel Oil (HSFO) (>0.5% by weight of Sulphur) in Emission Control Areas (ECA). There is a need for a more efficient means of predicting Sulphur content and differentiating between HSFO and Very Low Sulphur Fuel Oil (VLSFO) samples. This study compares the application of a Support Vector Machine (SVM) and Agglomerative Hierarchical Clustering (AHC) algorithm enhanced with Principal Component Analysis for dimensionality reduction purposes to predict HSFO and VLSFO marine fuel samples based on near-infrared (NIR) industrial data from North Sea operations correlated with laboratory-measured sulphur values instead of relying on lengthy laboratory-based measurements. The study also compares the effect of normalising the data by setting the area under the curve to one and standardising it by subtracting the mean of predictor variables and scaling by standard deviation. The results show that although >70% of HSFO samples were accurately predicted with the SVM, a better result was achieved using the unsupervised learning approach of AHC/PCA with >80% of HSFO samples correctly predicted despite the imbalance in the industrial data providing an effective model for the rapid and well-informed decision-making tool for vessel operators. Normalising the area under the curve to one produced similar results to using standardised data.

Keywords

Machine learning / Marine fuel / Support vector machines / Agglomerative hierarchical clustering / High sulphur fuel oil / Very low sulphur fuel oil

RIGHTS & PERMISSIONS

Harbin Engineering University and Springer-Verlag GmbH Germany, part of Springer Nature